Development of an adaptive online fuzzy arbitrator for forecasting short-term natural gas usage
Abstract
The focus of the work is on the development and utilization of a self-assembling Fuzzy logic controller for the purpose of improving short term natural gas load forecasts generated by artificial neural networks (ANN) and linear regression (LR) models. The approach is to form a matrix of dynamic post processors (DPP), composed of ARMAX models, which use load estimates generated by ANNs and LRs as inputs. The problem is to then determine the performance of each DPP under different operating conditions, and to generate a final load estimate using a Fuzzy logic controller. The contributions of this research are as follows. First, as part of a residuals analysis, prefiltering and nonlinear transforms are explored for the purpose of increasing the correlation of environmental input factors with gas load, while decreasing multicollinearity. This has the effect of reducing the covariance of model parameters and increasing forecast confidence. The result of this analysis will be used to develop ARMAX models to postfilter the ANN and LR forecast model estimates. The gas operating regions will be characterized by an adaptive clustering algorithm that will partition operating conditions into distinct patterns with unique consumption characteristics. Finally, an adaptive online Fuzzy controller identifies the characteristics of each DPP under different operating conditions, and generates a weighted average of the DPP estimators to produce the final gas load estimate.
This paper has been withdrawn.